A network-based approach to modelling bluetongue spread in France

A network-based approach to modelling bluetongue spread in France

Preventive Veterinary Medicine 170 (2019) 104744 Contents lists available at ScienceDirect Preventive Veterinary Medicine journal homepage: www.else...

2MB Sizes 0 Downloads 12 Views

Preventive Veterinary Medicine 170 (2019) 104744

Contents lists available at ScienceDirect

Preventive Veterinary Medicine journal homepage: www.elsevier.com/locate/prevetmed

A network-based approach to modelling bluetongue spread in France Noémie Courtejoie

a,b

b

a

, Simon Cauchemez , Gina Zanella , Benoît Durand

a,⁎

T

a

Epidemiology Unit, Laboratory for Animal Health, French Agency for Food, Environmental and Occupational Health and Safety (ANSES), University Paris-Est, 14 rue Pierre et Marie Curie, 94700 Maisons-Alfort, France Mathematical Modelling of Infectious Diseases Unit, Institut Pasteur, UMR2000, CNRS, 75015 Paris, France

b

A R T I C LE I N FO

A B S T R A C T

Keywords: Bluetongue Transmission dynamic modelling Contact network Host movement Vector-borne Vaccination

Bluetongue virus serotype 8 (BTV-8) was reported for the first time in Europe in 2006, causing the largest bluetongue outbreak ever recorded. France was mostly impacted in 2007/09. Trade restrictions were implemented all along. Vaccination became available from 2008: a limited number of doses was first administered in an emergency vaccination campaign, followed by two nationwide compulsory vaccination campaigns in 2009 and 2010. France regained a disease-free status in December 2012, but BTV may have kept circulating undetected as infected herds have been reported again since August 2015. We developed a stochastic dynamic compartmental model of BTV transmission in cattle and sheep to analyze the relative importance of vector active flight and host movements in disease spread, and assess the effectiveness of control measures. We represented BTV transmission both within and between French administrative subdivisions called cantons, during the 2007/ 09 outbreak and until the end of 2010, when compulsory vaccination was interrupted. Within-canton transmission was vector-borne, and between canton transmission could occur through three contact networks that accounted for movements of: (i) vectors between pastures located at close distance; (ii) cattle and sheep between pastures of the same farm; (iii) traded cattle. We estimated the model parameters by approximate Bayesian computation, using data from the 2007 French outbreak. With this framework, we were able to reproduce the BTV-8 epizootic wave. Host movements between distant pastures of the same farm were found to have a major contribution to BTV spread to disease-free areas, thus raising practical questions about herd management during outbreaks. We found that cattle trade restrictions had been well complied with; without them, the whole French territory would have been infected by winter 2007. The 2008 emergency vaccination campaign had little impact on disease spread as almost half vaccine doses had likely been administered to already immune cattle. Alternatively, establishing a vaccination buffer zone would have allowed a better control of BTV in 2008: limiting its spatial expansion and decreasing the number of infected cattle and sheep. We also showed a major role of compulsory vaccination in controlling the outbreak in 2009 and 2010, though we predicted a possible low-level circulation after the last detection.

1. Introduction Bluetongue (BT) is a non-zoonotic vector-borne viral disease of domestic and wild ruminants, mainly transmitted by biting midges of the genus Culicoides. Before the 21st century, BT incursions into Europe used to be sporadic and limited to the southern part of the continent (Mellor et al., 2008). Since 1998, they became more frequent and BT spread further North, hence becoming one of the most important diseases of livestock in Europe with strong economic and social consequences (Rushton and Lyons, 2015). Bluetongue virus serotype 8

(BTV-8) which was reported for the first time on the European continent in 2006, caused the largest BT outbreak ever recorded (Carpenter et al., 2009) with over 95,000 infected holdings detected in two years’ time. The strain that circulated in Europe surprised by its capacity to survive the coldest months and resume its spread after a winter break in a still poorly understood process referred to as overwintering. France was mostly impacted from 2007: BTV-8 progressed in an epizootic wave from North-East to South-West, crossing the country in two years ‘time. (Pioz et al., 2011). Trade restrictions were enforced in infected areas. An inactivated BTV-8 vaccine became available in spring

Abbreviations: AFSSA, French food safety agency; BTV(-8), bluetongue virus (serotype 8); CI95%, credible interval; CLC, CORINE land cover; CORINE, coordination of information on the environment ⁎ Corresponding author. E-mail addresses: [email protected] (N. Courtejoie), [email protected] (S. Cauchemez), [email protected] (G. Zanella), [email protected] (B. Durand). https://doi.org/10.1016/j.prevetmed.2019.104744 Received 1 April 2019; Received in revised form 1 July 2019; Accepted 8 August 2019 0167-5877/ © 2019 Elsevier B.V. All rights reserved.

Preventive Veterinary Medicine 170 (2019) 104744

N. Courtejoie, et al.

2008 in a limited amount. Vaccination was first voluntary; vaccine doses were released progressively and attributed preferentially to areas that had already been affected by BTV in the previous year to allow farmers to return to normal production conditions (Sénat, 2008). Then, two nationwide state-funded compulsory vaccination campaigns were implemented in the winters of 2008/09 and 2009/10. The outbreak died off and was considered to be over by December 2009. Vaccination became voluntary and self-funded in 2011 and 2012. It was banned from 2013 onwards to preserve the national bluetongue free status regained in December 2012. BTV-8 remained undetected in Europe until August 2015, when a strain with an almost identical genome sequence to the one that circulated in 2007/09 was detected in a ram in Central France (Bréard et al., 2016; Sailleau et al., 2017). The origin of the re-emergence remains unknown, with a possible silent circulation of BTV-8 in domestic ruminants between the two outbreaks (Courtejoie et al., 2017). Vaccination was re-introduced in autumn 2015. Knowledge gaps remain about the epidemiology and management of the unexpected 2007/09 outbreak, in particular on the following points: (i) burden of infection given the high proportion of asymptomatic animals; (ii) relative role of host and vector movements in disease spread; and (iii) effectiveness of control measures that were implemented vs alternative measures that could have been considered. In the past decades, mathematical models have been developed to study BT transmission and control in Europe (Courtejoie et al., 2018b). The challenging task of disentangling BTV spread via host and vector movements has rarely been addressed as many authors represented all routes of transmission together in a single probabilistic description (Szmaragd et al., 2009; Gubbins et al., 2010; de Koeijer et al., 2011; Boender et al., 2014; Bessell et al., 2016). Some authors explicitly considered long-distance host movements introduced by cattle trade (Turner et al., 2012; Ensoy et al., 2013; Sumner et al., 2017) but shortrange and non-commercial host movements were rarely accounted for. Here we developed a stochastic dynamic compartmental model of BTV spread in cattle and sheep from mainland France, representing long- and short-distance BTV transmission via three distinct contact networks explicitly accounting for different types of movements. The model was used to address remaining knowledge gaps on BTV spread and control.

Temperature data were obtained from the SAFRAN atmospheric analysis system maintained by Météo France, with a spatial resolution of 8 km. We extracted all daily temperatures from 2007 to 2010. Land cover data were extracted from the 2012 version of the CORINE (Coordination of information on the environment) Land Cover (CLC) database, provided by the European Environment Agency at a resolution of 100 m. Spatial data were aggregated by canton and temporal data were aggregated per week. Surveillance data consisted in the list of farms with confirmed clinical cases detected from July 2007 until December 2009. Confirmed clinical cases were defined as diseased animals showing BTV-8 clinical signs and for which BTV-8 genomes (or anti-BTV antibodies in early 2007 only) had been detected. These data were provided by the French Ministry for Agriculture and processed by Pioz et al. (2011). Serological data consisted in the results of a cross-sectional retrospective serological study conducted in winter 2007/08 in seven and four French departments for cattle and sheep respectively (Durand et al., 2010); a department is an administrative subdivision containing on average 36 cantons. The number of vaccines administered in each department during the 2008 emergency campaign and during the 2009 and 2010 nation-wide compulsory vaccination campaigns was provided by the French Ministry for Agriculture. 2.3. Model design and parametrization Stochastic compartmental models were used to capture BTV transmission in host populations in each canton. These models were operated with a weekly time step. Animals were grouped in species-specific compartments reflecting their health states (Fig. 1). We did not implement a compartmental representation of vector populations due to the absence of abundance data needed for model parametrization. No systematic Culicoides trapping was indeed performed prior to 2009 on the French territory, except in Corsica and along the Mediterranean coast (Baldet et al., 2004). We used a non-Markovian representation of BTV transmission between hosts to account for vector-borne transmission, and we integrated environmental-based proxies of vector abundance, survival and activity to account for the spatial and temporal variations of vector population dynamics. The size of cattle and sheep populations by canton was matched to real data. For cattle, we updated the number of animals and births per canton every week. For sheep, we assumed a constant size in each canton and applied a weekly renewal proportion (Supplement S2.A). At first, all animals of the canton were in the susceptible state. Ninf infected cattle were introduced in selected cantons: (i) on the observed date of first detection, the year when BTV-8 emerged (mid-July 2007); and (ii) at the beginning of each season of virus circulation (1st of June) afterwards. In 2007, infection was seeded in the six North-Eastern cantons where BTV-8 presence had first been confirmed. After 2007, the cantons where BTV was reintroduced in season n+1 were simulation- and season-specific: they were those where BTV was still circulating before the winter break in season n, that is on the date when temperatures dropped below the Tmin threshold in a proportion pow of cantons. Each week, the number of animals that became infected in a given sp (k , t ) ) was the sum of two terms: canton (ninf

2. Material and methods 2.1. Study area and study period We studied BTV-8 spread in mainland France (excluding overseas territories and Corsica) from summer 2007 to winter 2010, to cover the 2007/09 BTV-8 outbreak until the end of compulsory vaccination. We only focused on BTV-8, whereas BTV-1 circulated in the South of France in 2008 and 2009. We used administrative subdivisions called “cantons” as modelling units because sheep data were not available at smaller spatial scales. Each canton included on average 10 municipalities and covered about 150 km2. There were 3708 cantons in France during the study period; 3432 of them hosted cattle and/or sheep. 2.2. Data sources The number of cattle in each canton and all cattle movements between pairs of cantons were extracted from 2007 to 2010 from the National Identification Database, an exhaustive database maintained by the Ministry for Agriculture. The location and number of sheep in each canton were obtained from the 2010 Agriculture General Census of all holdings, conducted every ten years by the Ministry for Agriculture. We extracted pasture locations and the list of pastures belonging to the same farm from the Anonymized Land Registration System of 2011, provided by the French Agency for Services and Payment (Palisson et al., 2017). In this database, pastures were defined as grasslands, either permanent or temporary if part of a grass-arable rotation system.

sp sp sp ninf (k , t ) = n vect (k , t ) + nintro (k , t )

sp (k , t ) , the number of infectious animals introduced in canton k with nintro at time t, resulting from animal movements between trade partners or sp between distant pastures of the same farm; and n vect (k , t ) , the number of susceptible animals infected by vector bites. This latter number depends on the force of vector-borne infection from female midges located in the canton and in other cantons within flight distance. sp sp n vect (k , t ) ∼ Binom (S sp (k , t ) + Lsp (k , t ), Pinf (k , t ))

2

Preventive Veterinary Medicine 170 (2019) 104744

N. Courtejoie, et al.

Fig. 1. Schematic representation of the species-specific compartmental model. Superscripts c, s and sp are used to indicate parameter values specific to cattle, sheep or to either one of the species respectively. Cattle or sheep could be in one of the following health states: C, protected by colostral antibodies within their first months of life if born from seropositive mothers; S, susceptible, i.e. uninfected and immunologically naive; I, infectious, with enough BTV in the blood stream to infect Culicoides when feeding; R, recovered and protected against further infection by persistent antibodies; L, latentvaccinated, between vaccination and acquisition of protective immunity against BTV; V, vaccinated, for vaccinated individuals protected against BTV infection. The infectious health state was subdivided into msp stages so that the time spent in that state followed a flexible gamma distribution. We denoted: C sp (k , t ) , S sp (k , t ) , I sp (k , t ) , Rsp (k , t ) , Lsp (k , t ) and V sp (k , t ) , the number of animals of species sp in each health sp (k , t ) , the number of infectious state in canton k at time t; nintro animals of species sp introduced in canton k at time t. sp α1sp, α 2sp, Pinf (k , t ), v sp (k , t ), v2sp were the transition prob1

sp (k , t ) , probability of infection (vector-borne) in canton k at time t; abilities for species sp with: α1sp , 1/length of persistence of colostral antibodies; α 2sp , 1/viremia; Pinf v1sp, vaccination rate in canton k at time t; v2sp , rate of acquisition of protective immunity.

on arable lands and provide suitable breeding sites for BTV vector species (Ninio, 2011), whereas forests/semi-natural areas provide shelter to the wild animals that may contribute to BTV sylvatic cycle (Rossi et al., 2014). Between-canton movements of vectors and hosts occurred on three distinct contact networks: (i) the pasture network, representing midges flight; (ii) the farm network, representing movements of cattle or sheep between pastures of the same farm; and (iii) the trade network, representing movements of traded cattle. The nodes were cantons and a link existed between two cantons: (i) in the pasture network, if at least two pastures from each canton were less than one km apart, a distance used by Palisson et al. (2017) to represent the most likely routes of vector-borne disease transmission across the densely connected network of French pastures; (ii) in the farm network, if at least one farm had pastures located in each canton; (iii) in the trade network, if cattle had been traded between at least two farms located in each canton. The trade network was temporal and oriented, linking different donors and recipients every week, while the pasture and farm networks were static with links existing at all times and movements through these links as likely to go either way. Their topological properties are analyzed in Supplement S5. BTV transmission due to midges dispersal was represented by applying to canton k a fraction ΨP of the force of vector-borne infection of its neighbors on the pasture network (λvect(k,t)), where ΨP is the proportion of canton surface that can be reached by vectors coming from each neighboring canton. The number of infectious animals introduced sp (k , t ) ) depended on the through the farm and trade networks (nintro number of animals moved towards canton k on each network and on the prevalence of infection in the source canton. The total number of cattle traded could be fully informed by data, while the total number of cattle and sheep movements on the farm network depended on ΨF, the weekly proportion of animals moved between pastures of the same farm. Movements of traded cattle was subjected to restrictions that were implemented and complied with, with a probability θ. All these processes and associated parameters are described in more details in Supplement S1.B. Cantons with infected animals could be detected by passive clinical surveillance, given a probability Δ that infectious animals could show clinical signs and be detected (Supplement S2.B). Once at least one animal was detected, the canton became a “reporting canton”. We recorded the date of first detection per canton and applied similar control measures to those actually implemented during the outbreak: movements were banned in cantons located in a 20 km radius around the reporting ones; those located in a 90 km radius were placed in a

where the probability of infection of susceptible individuals is given by: sp Pinf

(k , t ) = 1 −

exp(−π sp *[λ

int

(k , t ) + λ vect (k , t )])

with π sp , the relative preference of vectors for cattle or sheep (conditional on feeding on these species); λint (k , t ) , the force of vector-borne infection from female midges located in canton k, and λ vect (k , t ) , the force of vector-borne infection from female midges located in other cantons within flight distance of canton k. λint (k , t ) represents the force of vector-borne infection from female midges located in the canton that got infected locally while feeding on infectious ruminants in the previous time steps, that completed the extrinsic incubation period (EIP) required for BTV replication and dissemination up to the arthropod vector salivary glands, and survived up to time t. We made the simplifying assumption that, in a given canton, and during the vector activity period, the vector to host ratio was constant. Under this assumption, the vector-borne transmission can be represented by a non-Markovian force of infection, which accounts for the Culicoides cohorts that emerged in the preceding weeks.

λint (k , t ) = τ (k , t )*∑ (wi * Prev (k , t − i)) i

with Prev (k , t − i) , the proportion of infectious animals at time t-i weighted by the species-specific trophic preferences of vectors; wi, the fraction of Culicoides vectors that have completed their EIP in i weeks and survived over that period (Supplement S1.A, C); τ (k , t ) , the weekly effective contact rate at which vectors and hosts from canton k come into effective contact, given by:

τ (k , t ) = β0 * Env (k )* b (k , t ) with β0, a coefficient that represents the baseline exposure of hosts to vectors, defined here as the product of the baseline vector to host ratio, the host to vector and vector to host probabilities of successful transmission, and the trophic preference of Culicoides for cattle and sheep vs other warm-blooded species; Env(k), the environmental variables used as proxy of host availability, Culicoides presence and abundance; b(k,t), the temperature dependent biting rate of Culicoides at time t in canton k that represents the seasonal variation in Culicoides activity. Env(k) was defined under the assumption that bluetongue transmission in a given area depends on the proportion of surface covered in pastures (CLC code: 231), where hosts and vectors come into contact. We used additional landscape metrics to modulate the transmission that occurred on pastures: the spatial density of borders between pastures and arable lands (CLC code: 211-213), and between pastures and forests/semi-natural areas (CLC code: 331-335). Indeed, manure is spread 3

Preventive Veterinary Medicine 170 (2019) 104744

N. Courtejoie, et al.

Table 1 Fixed parameter. Symbol

Description

α1c α2c

1/ length of persistence of colostral antibodies

0.0625 (1/16 wk)

(Vitour et al., 2011)

1/ viremia

0.25 (1/4 wk)

mc α1s

number of viremic stages in cattle 1/ length of persistence of colostral antibodies

3 0.07 (1/14 wk)

(Singer et al., 2001; Martinelle et al., 2011; Di Gialleonardo et al., 2011)

α2s ms πc

1/ viremia

0.33 (1/3 wk)

(Eschbaumer et al., 2010; Worwa et al., 2010)

number of viremic stages in sheep trophic preference for cattle vs sheep (if feeding on these species) biting rate (wk−1)

2 0.87

(Ayllón et al., 2014; Elbers and Meiswinkel, 2014)

b(k,t) μv EIP Tmin v2c

v2s ΨP** Δ

**

Ninf*** pow***

daily mortality proportion of Culicoides vectors extrinsic incubation period threshold temperature for virus replication weekly rate of acquisition of protective vaccinal immunity in cattle weekly rate of acquisition of protective vaccinal immunity in sheep proportion of canton surface reachable in a week by vectors from neighboring cantons probability of clinical onset and detection of infectious animals in newly infected areas number of infected cattle introduced to seed infection proportion of canton with Tp(k,t) < Tmin, used to model overwintering

Value

Reference

(Oura et al., 2010)

[0.00002 * Tp(k,t) * (Tp(k,t) - 3.7) * (41.9 – Tp(k,t))0.37]*7 6% (17-25 °C)

(Mullens et al., 2004)

11 days (17 °C) 12 °C 0.35 (t vc = 7 wk)*

(Carpenter et al., 2011) (Carpenter et al., 2011) (Merial, BTVPUR®, AlSap8)

0.52 (t vs = 4 wk)*

(Intervet, BOVILIS BTV8 ®)

0.4

Flight distances (Kluiters et al., 2015), cantons surface (Supplement S2.C) (Durand et al., 2010; Mounaix et al., 2010; Courtejoie et al., 2018a) (Supplement S2.B)

0.02

(Goffredo et al., 2004)

5 90%

c for cattle, s for sheep, v for vectors; wk, weeks; Tp(k,t), temperature in canton k at time t. sp * so that (1 − (1 − v2sp)tv ) = 95% , with t vsp the time before reaching immunity in 95% of the vaccinated animals. ** varied in a sensitivity analysis on parameter estimates. *** varied in a sensitivity analysis on model predictions.

the species-specific number of seropositive animals detected in each department sampled in the serosurvey conducted in winter 2007/08 (Supplement S4.B). We used uniform priors for all parameters (Supplement S4.C). We investigated the need to make within-canton transmission rates vary with land-cover metrics, and the need for between-canton transmission to occur through only one or several contact networks. We built separate models including various combinations of the variables and contact networks of interest and compared them using a model selection procedure based on random forest classification methods (Pudlo et al., 2016). We selected the set of variables/networks providing the best fit to the observed data, then used it for all subsequent analyses (Supplement S3).

restricted zone: movements were allowed within that zone, but prohibited from the inside to the outside. We represented three vaccination campaigns: the 2008 emergency vaccination campaign, conducted in times of outbreak, and the 2009 and 2010 compulsory campaigns, conducted in the first months of each year, when vectors were not active. In 2008, we attributed the limited number of vaccine doses following the Ministry for agriculture’s vaccination schedule (Fig. 6.A, D). Most model parameters were informed from the literature, or from plausible assumptions then challenged in sensitivity analyses (Table 1). Three of them were estimated because they were specific to our study context and could not be inferred from previous studies. 2.4. Parameter estimation and model selection

2.5. Model implementation, validation and exploitation We estimated three parameters: β0, the baseline exposure of hosts to vectors; ΨF, the proportion of animals moved weekly between pastures of the same farm; and θ, the probability that cattle trade control measures would be complied with. We used the Adaptive population MonteCarlo approximate Bayesian computation method (ABC-APMC) (Lenormand et al., 2013), a likelihood-free method useful for complex, stochastic models where the full likelihood cannot be estimated. It is based on the generation of joint parameters values (particles) initially sampled from the joint prior distribution of each parameter, followed by the selection of the particles for which the model outputs (summary statistics) satisfy a proximity criterion with the target data (Supplement S4.A). We used the following settings: 0.5 for the quantile of the distribution of distances to observed data used to define tolerance thresholds; 0.03 for the minimal proportion of new particles satisfying the stopping criteria from the previous step; and a final size of 5000 particles used to build posterior probabilities. The summary statistics used for inference were built from surveillance and seroprevalence data from the 2007 epizootic wave. For surveillance data, we used the numbers of departments with, and without, reporting cantons by winter 2007/08. For seroprevalence data, we used

The model was coded in C++ and operated in R (version 3.3.2) using the Rccp package. ABC-APMC estimation and model comparison by random forest were conducted using the EasyABC and ABC-RF packages in R. To assess the ability of our framework to estimate parameter values using the chosen summary statistics, we simulated 100 epidemics with parameters randomly drawn from the prior distributions, and we estimated back these parameters using the ABC-APMC procedure (Supplement S6). Model fit was evaluated by sampling 1000 particles from the weighted joint posterior distributions and by generating summary statistics that we compared to the observed ones (used for parameter estimation). An external validation was performed by confronting simulated data with the observed spatio-temporal distributions of reporting cantons from 2007 to 2010 (not used for parameter estimation). From 2008, we excluded the southern areas where BTV-1 circulated as there may have been some cross-immunity. A sensitivity analysis was performed to evaluate the effect on the estimated parameter values of two key parameters with values that were fixed: the proportion of canton surface reachable by vectors from 4

Preventive Veterinary Medicine 170 (2019) 104744

N. Courtejoie, et al.

Fig. 2. Posterior distributions of the three estimated parameters. A. Baseline exposure of hosts to vectors (β0); B. Proportion of animals moved weekly through the farm network (ΨF); C. Probability of control measures being implemented on movements of cattle through the trade network (θ). CI95%: credible interval.

3. Results

neighboring cantons (ΨP) and the probability of detection upon clinical suspicion (Δ). We compared (i) the relative error induced by a 25% change of each fixed parameters on the average values of each estimated parameter, with (ii) the coefficient of variation of the posterior distributions obtained with the default values. In addition, we investigated the effect of fixed deviations of initial conditions (Ninf, pow) on model predictions (Supplement S7). We operated the parametrized model until the end of 2010, using 1000 particles sampled from the weighted joint posterior distributions, and computed various indicators. To address the epidemiological contribution of the contact networks during the 2007 and 2008 epizootic waves, we investigated the proportion of transmission that occurred through each of them. In every simulation, we recorded the source of infection of each newly infected canton, i.e. whether a canton previously free of infection had been contaminated through the pasture, farm or trade network. Infections that occurred through multiple networks on the same week were randomly allocated to either one of them. To address the true burden of infection, detected or not, and to highlight local differences in the extent of BTV spread, we reconstructed for all French departments: (i) the seroprevalence level in the winter after each season of virus circulation (2007, 2008, 2009 and 2010); and (ii) the cumulative proportion of animals that had been infected in each season of virus circulation. To evaluate the control measures, we estimated the proportion and number of vaccines that had been administered to already immune animals in the 2008 emergency vaccination campaign. Finally, we explored alternative control scenarios. We investigated four alternative scenarios of movement restriction in 2007: one in which no control measures were applied on trade movements of cattle, two in which they were applied and complied with at 90% and 95%, and one in which movement restrictions, perfectly complied with, were extended to movements of animals between pastures of the same farm. We investigated two alternative scenarios of vaccination from 2008: one in which there was no vaccination at all, neither in 2008, nor in the compulsory campaigns of 2009 and 2010; and another one in which the 2008 emergency vaccination campaign was targeted to create a buffer zone beyond the previously affected areas (Figure S1), as recommended by the French food safety agency at that time (AFSSA, 2008). In the latter scenario called the “AFSSA scenario”, we released the same number of doses every week as in the baseline scenario, as vaccines were limiting at the time, but we distributed them in different order of priority, vaccinating less areas but with higher vaccination rates. We ran 1000 simulations in each scenario.

3.1. Description of the study area The study area comprised the 3432 French cantons that hosted cattle or sheep in 2007/10. There was a total of 19.6 million head of cattle and 5.5 million head of sheep hosted in 236 and 55 thousand farms, respectively. These domestic ruminants may have been put out to pasture on the three million parcels of grasslands defined as pastures (of 0.05 km2 on average). The median number of cattle and sheep per canton was 3042 [1st – 3rd quartile: 606 – 8,715] and 347 [80–1135], respectively; and the median number of cattle and sheep farms per canton was 45 [15–92] and 8 [2–19], respectively. The median number of pastures was 573 [137–1,299] per canton, 7 [3–14] per farm and 9 [4–15] per farm with more than one pasture (i.e. 90% of all farms). 33% of all farms, and 37% of those with more than one pasture had pastures located in different cantons.

3.2. Model selection and parameter estimation Model selection showed: (i) that the proportion of pastures was crucial to representing BTV within-canton transmission, with no benefit to model fit when including additional landscape metrics (Supplement S3.B); and (ii) that no network on its own was enough to represent BTV spread to new areas, with the best fit obtained when all networks were combined (Supplement S3.C). We thus selected the model in which the only environmental variable (Env(k)) was the proportion of canton surface covered in pastures, and which included the three contact networks. The framework and choice of summary statistics were validated based on pseudo-observations generated from randomly chosen parameter values (Supplement S6). Parameter estimates appeared satisfactory but estimates were regressed towards the mean of the prior distribution for extreme parameter values because of saturation in the summary statistics. Then, we applied the framework to the observed data (Fig. 2). The posterior distributions had the following median values: 5543 (CI95%: 3,078–9,340) for the baseline exposure of hosts to vectors (β0); 60.4% (CI95%: 27.4–96.0%) for the proportion of animals moved weekly between pastures of the same farm (ΨF); and 97.1% (CI95%: 92.0–99.7%) for the probability that control measures would be complied with (θ). Simulated data allowed reconstructing the observed data used for parameter estimation (Figure S3), as well as the spatio-temporal distribution of reporting cantons (Fig. 3). As in the observations, the simulations predicted a peak in detections in 2007, followed by a winter break and a second peak in 2008 when virus circulation resumed (Fig. 3.B). The ability of the model to reconstruct the epizootic wave 5

Preventive Veterinary Medicine 170 (2019) 104744

N. Courtejoie, et al.

Fig. 3. External validation: spatio-temporal pattern of the apparent infection. A. Spatial pattern of detection of infection in cantons: frequency of reconstructed detection (1000 simulations: A.1, A.3) vs observations (A.2, A.4), in 2007 (A.1, A.2) and in 2008 (A.3, A.4); B. Temporal pattern of detection: histogram of observed and simulated reporting cantons in 2007/10 (median value of 1000 simulations).

(Fig. 5.B3-4, D.3-4), which would have been detected in our setting (Figure S5), though the outbreak was considered as over from 2010 onwards. In 2008, vaccination was conducted during the season of virus circulation. In our simulations, we highlighted spatial contrasts in the proportion of vaccines that had been administered to already immune animals (Fig. 6.C, F) due to the relative timing of vaccination and infection (Fig. 6.A, D). For both species, most of the vaccinated animals in the North-Eastern departments were already immune (> 80% in some areas) and we estimated that > 3 million vaccine doses had been administered to already immune cattle (41% of all vaccines), and < 1 million (18%) to already immune sheep. In 2008, vaccination, as it was conducted, had little impact on spatial spread in our simulations. The absence of vaccination would have only resulted in a 5% increase in the number of newly reporting cantons (Fig. 7.B). However, there would have been a greater increase in the number of infected animals (about 10% increase in cattle and 55% in sheep) (Fig. 7.A). The alternative AFSSA scenario would have allowed an additional 15% reduction in the number of newly reporting cantons, and an additional 20% and 30% reduction in the number of infected cattle and sheep respectively. However, the infected cases would have been distributed differently than in the baseline scenario, with more cases in the North-East and less in the South-West (Fig. 7.D). Overall, less vaccine doses would have been administered to immune animals, with only 5% reduction in the number of useless doses in sheep (0.8 vs 0.9 million) but over 60% reduction in cattle (1.2 vs 3.2 million). Finally, we predicted that from 2009, the absence of vaccination would have led to a dramatic increase in the number of infected animals in both 2009 and 2010, even greater in sheep than in cattle (Fig. 7.A). If movements on the farm network had been controlled similarly to the ones on the trade network in 2007, there would have been a 40% decrease in the number of newly reporting cantons compared to the baseline scenario (Fig. 7.B, C.1), as well as a 40% decrease in the number of infected animals in that year (both in cattle and sheep) (Fig. 7.A). On the other hand, if movements on the trade network had not been controlled in 2007, > 65% of the French cantons would have reported BTV-8 infected cases by winter 2007 (Fig. 7.B), > 100% more than in the baseline scenario (Fig. 7.C.2). There would have been a dramatic increase in the number of infected cattle and sheep (250 and 300% respectively, Fig. 7.A), meaning that > 70% and > 45% of the total cattle and sheep populations respectively would have been infected. The effect would have been less dramatic but still substantial with smaller decreases in the compliance of movement restriction (Fig. 7.A, B). The sensitivity analysis showed little effect on parameter estimates of a 25% variation of the probability of detection upon clinical suspicion (Δ), but a stronger effect of a 25% variation of the proportion of

that crossed France in 2007 and 2008 was illustrated by mapping the newly reporting cantons every six weeks (Figure S4). In 2007, the map (Fig. 3.A) and histogram (Fig. 3.B) of reporting cantons showed slightly more notifications on average in simulations vs observations. By the end of winter, the area with reporting cantons in most simulations matched the area with most observed reporting cantons: apparent infection was mostly limited to the North-East of the country (Fig. 3.A.1-2). Yet, in a few simulations, BT cases could have been detected in the whole territory during the 2007 epizootic wave (Fig. 3.A.1). The 2008 epizootic wave progressed towards the South-West, reaching similar geographical areas in simulations and observations (Fig. 3.A.3-4). In both case, two years of BTV circulation resulted in infected cases detected in > 95% of the French departments (excluding those where BTV-1 circulated). In 2009, BTV kept circulating in the already detected areas, with no observed newly reporting cantons (Fig. 3.B). However, in simulations, infection spread slightly further South-East in 2009, hence the few newly reporting cantons (Fig. 3.B). 3.3. Model exploitation In our simulations, most transmission to new areas occurred on the farm network (65%), followed by the pasture network (35%), and very little from trade (< 1%) (Fig. 4). In 2007, the reconstructed seroprevalence levels (Fig. 5.A.1, C.1) and cumulative proportion of infected animals per department (Fig. 5.B.1, D.1) conveyed the same information: the burden of infection. They highlighted spatial contrasts, with some areas where more than 90% of the ruminants may have been infected by winter 2007/08. These maps diverged from 2008 (Fig. 5.A-D.2-4), as several processes contributed to seroprevalence: past and present infection, population renewal, and vaccination. The contrasts between these maps gave an indication of the relative contribution of these processes. We predicted that BTV was still circulating in 2010 with similar low levels as in 2009

Fig. 4. Initial source of infection. Proportion of BTV introduction to new areas that happened on each contact network (pasture, farm or trade networks). The boxplots indicate the mean, interquatile interval, minimal and maximal values. 6

Preventive Veterinary Medicine 170 (2019) 104744

N. Courtejoie, et al.

Fig. 5. Reconstructed seroprevalences and proportions of livestock infected per department, from 2007 to 2010. A, C. Seroprevalences (due to natural infection or to vaccination) after each season of virus circulation in cattle (A) and sheep (C); B, D. Cumulative proportion of animals infected in each season of virus circulation in cattle (B) and sheep (D) ; 1-4. season of virus circulation: 2007 (1), 2008 (2), 2009 (3) and 2010 (4).

(Supplement S7.B).

canton surface reachable by vectors coming from neighboring cantons (ΨP) (Supplement S7.A). However, we showed little difference on the variation of model predictions for each couple of ΨP and associated parameter estimates. Lastly, the sensitivity analysis on model predictions showed little effect of variations of the initial conditions

4. Discussion In this work, we developed a stochastic dynamic model of Fig. 6. Evaluation of the 2008 emergency vaccination campaign per department. A, D. Vaccination schedule: order of priority for the distribution of the limited number of vaccine doses spread out between May and September 2008 (DGAL/SDSPA, 2008), in cattle (A) and sheep (D), the order of priority is indicated by the color code; B, E. Vaccination coverage achieved by the end of the campaign (October 2008) in cattle (B) and sheep (E); C, F. Proportion of vaccine doses administered to already immune animals in cattle (C) and sheep (F).

7

Preventive Veterinary Medicine 170 (2019) 104744

N. Courtejoie, et al.

Fig. 7. Impact of the alternative control scenarios on spatial spread and outbreak size. A, B. Country-wide percentage of variation compared to the baseline scenario of alternative measures of movement restrictions and of alternative vaccination strategies: number of infected cattle and sheep from 2007 to 2010 (A); number of cantons first detected in 2007 and 2008 (B). C. Simulated variation of the frequency of canton detection in 2007 (1000 simulations) in two alternative scenarios of movement control: with additional control of movements between pastures of the same farm (C.1), with no control of trade movements (C.2). D. Variation of the number of infected cattle in 2008, per department, in the AFSSA vaccination scenario: in cattle (D.1) and sheep (D.2).

cantons. Most transmission events between cantons were predicted to have happened on the farm network. Movements between distant pastures of the same farm are rarely considered in bluetongue transmission models because they are poorly documented. There is no precise record of grazing practices that may vary across geographical areas, breeding types and farmers. Our parameter estimation meant that grazing ruminants changed pasture on average every two weeks, which seems consistent given that French pastures are small (0.05 km2 on average) and that animals are frequently moved for grass renewal and sanitary reasons such as the interruption of parasitic cycles (e.g. Fasciola hepatica). The major contribution of movements between distant pastures of the same farm leads to practical implications as we showed in a simulation study that controlling these movements may have prevented many infections and limited the geographical spread. However, these findings raise crucial questions about the feasibility of such control measures and about management practices as grazing habits are at the discretion of farmers. Movements between distant pastures of the same

bluetongue transmission in French cattle and sheep. We represented BTV vector-borne transmission in infected cantons, and used contact networks to represent BTV spread to disease-free areas. Our framework had the specificity of integrating two types of host movements: cattle traded between farms and cattle and sheep moved between distant pastures of the same farm. We combined multiple and high quality data sources to represent exhaustively population dynamics processes in hosts. Because of the absence of such data for Culicoides during the study period, we represented BTV vector-borne transmission in infected cantons by a non-Markovian formulation of the force of infection. This is equivalent to using a compartmental representation of vector populations with a fully Markovian dynamics, assuming that the vector to host ratio remains constant during the vector activity period (cantonand year-specific). This model may be adapted to the study of other vector-borne diseases of ruminants, in areas where the vector abundance does not show strong variations during the vector activity period. We used our model to address the question of the relative contribution of the contact networks to disease spread between French 8

Preventive Veterinary Medicine 170 (2019) 104744

N. Courtejoie, et al.

record of the number of traded sheep in France. However, there are over four times more cattle farms than sheep farms in France, and there is little live sheep trade as most animals only leave their birth farm when sent to the slaughter house. In addition, the movements we used were those that effectively took place as movements remained possible under specific protocols (1266/2007/CE, October 26 2007). We may under-estimate the movements that would have happened in 2007/09 in absence of outbreak, as movement restrictions are likely to have impacted: (i) the number of sales, with a 21% decrease estimated in a beef cattle breed in which most calves are sold for fattening (Tago et al., 2014); (ii) export destination, in relation to the evolution of restricted zones; and (iii) timing, as practical constraints add export delays. However, the analysis of the French cattle trade network from 2005 to 2009 did not show any significant difference between the years of the study period at the national scale (Dutta et al., 2014). We also retrospectively investigated the usefulness of vaccination. The 2008 emergency vaccination campaign did not prevent disease expansion to new areas. Vaccination targeted in priority the NorthEastern departments where most animals had already been infected in 2007, so that most vaccine doses were administered to already immune animals. Vaccination targeted ahead of the front would have limited BTV spatial spread: it would have been preventive in the areas that had not been reached by the epizootic wave in 2007, and the 2008 outbreak would have remained constrained to the already infected areas. Back in 2008, the design of the vaccination strategy had been individual-based, to protect the farmers that had already suffered from the 2007 epizootic wave, rather than population-based, to prevent disease expansion to new areas. Here we show that both vaccination strategies, individual or population-based, have –or would have- met their respective goals. Whatever the vaccination strategy, only vaccination performed by July 2008 may have provided protective immunity on time and influenced the course of the outbreak as: (i) vaccination was conducted simultaneously with virus circulation, and (ii) there is a few weeks’ delay between vaccination and acquisition of protective immunity. Vaccination was more preventive in sheep than in cattle because it started earlier in this species, which is more sensitive to BTV, and because only one vaccine dose was required in sheep vs two in cattle. Vaccination became truly preventive from 2009, when vaccines became available in sufficient quantity to vaccinate all domestic ruminants outside of the periods of vector activity. The course of the outbreak was truly changed by widespread compulsory vaccination which allowed maintaining high seroprevalence levels. Without vaccination, BTV would have kept reemerging every year with a significant level of infected cattle and sheep, suggesting that the situation may have become endemic. This is consistent with the results obtained by the EFSA Panel on Animal Health and Welfare (EFSA, 2017) whose mathematical model indicated that BTV could persist for several years without any vaccination, reaching an endemic situation with low level of prevalence of infection (1.5% in cattle, 0.6% in sheep). We still do not know whether widespread compulsory vaccination allowed a real eradication of BTV as our model predicted a potential residual level of circulation even after the last case detection in 2009. The model developed by EFSA experts predicts that five years of vaccination of 95% of susceptible French cattle and sheep would have been required to reach a prevalence of infection close to eradication levels (EFSA, 2017). If vaccination went on after 2010, it became voluntary and there is little knowledge on vaccine uptake at that time. We suggested in a previous study that vaccination have been only little implemented, even less in 2012 than in 2011 (Courtejoie et al., 2018a). It would be interesting to model alternative vaccination scenarios after 2011, such as one or two additional compulsory campaigns, and assess whether the 2015 re-emergence could have been prevented. Some of our modelling assumptions need to be discussed. The resurgence after each winter break was obtained providing assumptions on BTV overwintering, a phenomenon that remains poorly understood and most likely results in the combination of several processes

farm are also harder to regulate than trade exchanges, which are the subject of specific protocols and are rigorously traced. Among the many mathematical models of BTV-8 transmission developed after the European outbreak (Courtejoie et al., 2018b), Sumner et al. (2017) were the only ones providing a thorough quantification of the relative contribution of host and vector movements to transmission events between farms. They attributed > 90% of between-farm transmission to vector-dispersal, which does not contradict our results given that the epidemiological units are different (farms vs cantons), and that within-canton transmission is mainly driven by vectors in our model. Here we provide an additional layer of information about the drivers of BTV spread as we focus on the role played by different types of contact networks in BTV spread to new areas at the wider scale of the canton, with a median number of 94 farms per canton. In previous modeling studies, a greater attention has been paid to the diversity of vector dispersal modes (e.g. active, passive, against the wind, Hendrickx et al., 2008; Ducheyne et al., 2011; Sedda et al., 2012) than to the diversity of host movements. When the latter were explicitly represented, only long-range movements of traded hosts were accounted for (Turner et al., 2012; Ensoy et al., 2013; Sumner et al., 2017), and not non-commercial animal movements that may happen at a similar distance to that of vector active flight. Only Graesbøll et al. (2012; 2014) provided a detailed representation of both host and vector-related processes in BTV transmission: in vectors, they separated active flight and passive wind-borne dispersal; in hosts, they represented the movements of animals between pastures of the same farm under four different grazing conditions, as well as the mixing of animals from neighboring farms. Yet, their highly detailed framework did not allow quantifying the relative contribution of short-range active flight by midges and movements of hosts on pasture because of the high sensitivity of their model to parameter values, the poor knowledge on the flying parameters, and the lack of data on both host and vector distributions (Graesbøll et al., 2012). We provided a simpler representation of transmission processes designed to best use available data and existing literature. We did not describe several modes of vector-borne dispersal, but considered that, in addition to being responsible for BTV spread inside cantons, vectors could spread infection between cantons through the pasture network, with flight distance as only limiting criteria. It is possible that part of the transmission that we attribute to host movements between pastures was actually due to vector active dispersal at a wider scale than that considered (> 5 km per week), or to passive wind-borne vector dispersal. On the other hand, part of the transmission attributed to vector dispersal in previous studies may be due to non-commercial host movements. Yet, results obtained for BTV spread in French livestock may differ in other European countries as grazing habits depend on breeding types and country-specific management practices. Movements of traded cattle were the only ones allowing for longdistance jumps and fast spreading. However, they hardly contributed to disease spread in simulations because the control measures were almost perfectly implemented and complied with. The analysis of alternative control scenarios stressed on the need for an efficient control of trade movements in times of outbreak, as we predicted a dramatic increase on both BTV spatial spread and outbreak size if they were only 5% less controlled, and a possible infection of the whole French territory by the end of 2007. Animal transport restrictions had already been proven effective in substantially slowing down BTV spatial spread in Europe in 2006 and 2007 (de Koeijer et al., 2011; Boender et al., 2014), and in reducing outbreak sizes in Belgium in 2006 (Ensoy et al., 2013) and in Eastern England in 2007 (Turner et al., 2012). Furthermore, when BTV was introduced in the UK in 2007, the movement restrictions already in place as a result of foot-and-mouth disease control were identified as one of the main factors explaining the relatively small 2007 outbreak in the UK, compared with other European countries (Turner et al., 2019). Our analysis of host movements was limited by available data. We represented only cattle trade movements as there is no comprehensive 9

Preventive Veterinary Medicine 170 (2019) 104744

N. Courtejoie, et al.

Appendix A. Supplementary data

(Takamatsu et al., 2004; Napp et al., 2011). It may be explained by the persistence of adult vectors in the coldest months by taking shelter inside farm buildings (Baldet et al., 2008; Carpenter et al., 2009); or by vertical transmission in hosts, with a cumulative duration of infectious viremia in heifers and calves lasting longer than the vector inactivity period (Wilson et al., 2008). Our model does not have the granularity allowing to represent BTV overwintering, but we assumed that BTV resumed its spread in the cantons where it was still circulating when temperatures dropped in the end of each season of circulation. The sensitivity analysis showed little variation of model predictions with reasonable variations of the initial conditions used for BTV reintroduction. The surveillance system was based on clinical suspicion and we used a single probability of detection of infected animals upon clinical suspicion, though this may have varied in time, in place, according to the main breeding type and to the sensitization of farmers. In 2010, we predicted a low-level virus circulation, which would have been detected if applying the same probability of detection. Yet, no case had been detected in this year. We may have re-seeded infection too strongly in 2010: our assumption for overwintering may not be adapted to the epidemiological context after 2009 when there was no, or low-level, virus circulation. On the other hand, the probability of detecting animals upon clinical suspicion may have decreased in time, allowing for an undetected low-level BTV circulation in 2010 and potentially up to the 2015 reemergence. Indeed, the 2015 BTV-8 strain, though almost genetically identical to the one isolated in 2007, has been shown to induce less clinical signs in sheep experimentally infected with both strains (Flannery et al., 2019). In conclusion, we built a framework that allowed the reconstruction of the 2007/09 BTV outbreak in France. We showed a major contribution to BTV spread between cantons of host movements between distant pastures of the same farm, raising practical questions of herd management in times of outbreak. We provided an assessment of the effectiveness of the control measures that had been conducted, stressing on the crucial impact of the restriction of cattle trade movements, and providing a better understanding of the impact of the successive vaccination campaigns until the outbreak died off. This adaptable framework could be further used to reproduce and understand past events such as the cumulative impact of vaccination and population renewal in shaping the immunity landscape in French ruminants until the 2015 reemergence. In the future, this framework might become a management tool to explore and compare various control scenarios in times of outbreak.

Supplementary material related to this article can be found, in the online version, at doi:https://doi.org/10.1016/j.prevetmed.2019. 104744. References AFSSA, 2008. Avis de l’Agence française de sécurité sanitaire des aliments sur le risque de diffusion de la fièvre catarrhale ovine à sérotypes 1 et 8 en France et les mesures associées pour en diminuer le niveau. Saisine 2008-SA-0033. . Ayllón, T., Nijhof, A.M., Weiher, W., Bauer, B., Allène, X., Clausen, P.-H., 2014. Feeding behaviour of Culicoides spp. (Diptera: Ceratopogonidae) on cattle and sheep in northeast Germany. Parasit. Vectors 7, 34. https://doi.org/10.1186/1756-3305-7-34. Baldet, T., Delécolle, J.C., Mathieu, B., de La Rocque, S., Roger, F., 2004. Entomological surveillance of bluetongue in France in 2002. Vet. Ital. 40, 226–231. Baldet, T., Delécolle, J.C., Cêtre-Sossah, C., Mathieu, B., Meiswinkel, R., Gerbier, G., 2008. Indoor activity of Culicoides associated with livestock in the bluetongue virus (BTV) affected region of northern France during autumn 2006. Prev. Vet. Med. 87, 84–97. https://doi.org/10.1016/j.prevetmed.2008.06.014. Bessell, P.R., Searle, K.R., Auty, H.K., Handel, I.G., Purse, B.V., Bronsvoort, B.M., de, C., 2016. Assessing the potential for Bluetongue virus 8 to spread and vaccination strategies in Scotland. Sci. Rep. 6, 38940. https://doi.org/10.1038/srep38940. Boender, G.J., Hagenaars, T.J., Elbers, A.R.W., Gethmann, J.M., Meroc, E., Guis, H., de Koeijer, A.A., 2014. Confirmation of spatial patterns and temperature effects in Bluetongue virus serotype-8 transmission in NW-Europe from the 2007 reported case data. Vet. Res. 45, 75. https://doi.org/10.1186/s13567-014-0075-x. Bréard, E., Sailleau, C., Quenault, H., Lucas, P., Viarouge, C., Touzain, F., Fablet, A., Vitour, D., Attoui, H., Zientara, S., Blanchard, Y., 2016. Complete genome sequence of bluetongue virus serotype 8, which reemerged in France in August 2015. Genome Announc. 4. https://doi.org/10.1128/genomeA.00163-16. Carpenter, S., Wilson, A., Barber, J., Veronesi, E., Mellor, P., Venter, G., Gubbins, S., 2011. Temperature dependence of the extrinsic incubation period of orbiviruses in Culicoides biting midges. PLoS One 6, e27987. https://doi.org/10.1371/journal.pone. 0027987. Carpenter, S., Wilson, A., Mellor, P.S., 2009. Culicoides and the emergence of bluetongue virus in northern Europe. Trends Microbiol. 17, 172–178. https://doi.org/10.1016/j. tim.2009.01.001. Courtejoie, N., Durand, B., Bournez, L., Gorlier, A., Bréard, E., Sailleau, C., Vitour, D., Zientara, S., Baurier, F., Gourmelen, C., Benoit, F., Achour, H., Milard, C., Poliak, S., Pagneux, C., Viarouge, C., Zanella, G., 2017. Circulation of bluetongue virus 8 in French cattle, before and after the re-emergence in 2015. Transbound. Emerg. Dis. https://doi.org/10.1111/tbed.12652. Courtejoie, N., Salje, H., Durand, B., Zanella, G., Cauchemez, S., 2018a. Using serological studies to reconstruct the history of bluetongue epidemic in French cattle under successive vaccination campaigns. Epidemics. https://doi.org/10.1016/j.epidem. 2018.05.005. Courtejoie, N., Zanella, G., Durand, B., 2018b. Bluetongue transmission and control in Europe: a systematic review of compartmental mathematical models. Prev. Vet. Med. 156, 113–125. https://doi.org/10.1016/j.prevetmed.2018.05.012. de Koeijer, A.A., Boender, G.J., Nodelijk, G., Staubach, C., Meroc, E., Elbers, A.R.W., 2011. Quantitative analysis of transmission parameters for bluetongue virus serotype 8 in Western Europe in 2006. Vet. Res. 42, 53. https://doi.org/10.1186/1297-971642-53. DGAL/SDSPA, 2008. Note d’information : Fièvre Catarrhale Ovine – Stratégie Vaccinale BTV 8. Di Gialleonardo, L., Migliaccio, P., Teodori, L., Savini, G., 2011. The length of BTV-8 viraemia in cattle according to infection doses and diagnostic techniques. Res. Vet. Sci. 91, 316–320. https://doi.org/10.1016/j.rvsc.2010.12.017. Ducheyne, E., Lange, M., Van der Stede, Y., Meroc, E., Durand, B., Hendrickx, G., 2011. A stochastic predictive model for the natural spread of bluetongue. Prev. Vet. Med. 99, 48–59. https://doi.org/10.1016/j.prevetmed.2011.01.003. Durand, B., Zanella, G., Biteau-Coroller, F., Locatelli, C., Baurier, F., Simon, C., Le Dréan, E., Delaval, J., Prengère, E., Beauté, V., Guis, H., 2010. Anatomy of bluetongue virus serotype 8 epizootic wave, France, 2007-2008. Emerg. Infect. Dis. 16, 1861–1868. https://doi.org/10.3201/eid1612.100412. Dutta, B.L., Ezanno, P., Vergu, E., 2014. Characteristics of the spatio-temporal network of cattle movements in France over a 5-year period. Prev. Vet. Med. 117, 79–94. EFSA AHAW Panel (EFSA Panel on Animal Health and Welfare), 2017. Scientific opinion on bluetongue: control, surveillance and safe movement of animals. EFSA J. 15 (3), 4698. https://doi.org/10.2903/j.efsa.2017.4698. Elbers, A.R.W., Meiswinkel, R., 2014. Culicoides (Diptera: Ceratopogonidae) host preferences and biting rates in the Netherlands: comparing cattle, sheep and the blacklight suction trap. Vet. Parasitol. 205, 330–337. https://doi.org/10.1016/j.vetpar. 2014.06.004. Ensoy, C., Aerts, M., Welby, S., Van der Stede, Y., Faes, C., 2013. A dynamic spatiotemporal model to investigate the effect of cattle movements on the spread of bluetongue BTV-8 in Belgium. PLoS One 8, e78591. https://doi.org/10.1371/journal. pone.0078591. Eschbaumer, M., Wäckerlin, R., Rudolf, M., Keller, M., König, P., Zemke, J., Hoffmann, B., Beer, M., 2010. Infectious blood or culture-grown virus: a comparison of bluetongue virus challenge models. Vet. Microbiol. 146, 150–154. https://doi.org/10.1016/j. vetmic.2010.05.004.

Funding sources This work was supported by the Ile-de-France Region as part of the DIM1Health project. SC acknowledges financial support from the AXA Research Fund, the Investissement d’Avenir program, the Laboratoire d’Excellence Integrative Biology of Emerging Infectious Diseases program (Grant ANR-10-LABX-62-IBEID), the Models of Infectious Disease Agent Study of the National Institute of General Medical Sciences and the INCEPTION project (PIA/ANR-16-CONV-0005).

Declaration of Competing Interest Co-authors do not have any competing interest.

Acknowledgements We thank all data providers, especially the French Ministry for agriculture, the departmental veterinary laboratories and the French national reference laboratory for bluetongue disease (Corinne Sailleau, Emmanuel Bréard, Stéphan Zientara) and Maryline Pioz (INRA). 10

Preventive Veterinary Medicine 170 (2019) 104744

N. Courtejoie, et al.

Pioz, M., Guis, H., Calavas, D., Durand, B., Abrial, D., Ducrot, C., 2011. Estimating frontwave velocity of infectious diseases: a simple, efficient method applied to bluetongue. Vet. Res. 42, 60. https://doi.org/10.1186/1297-9716-42-60. Pudlo, P., Marin, J.-M., Estoup, A., Cornuet, J.-M., Gautier, M., Robert, C.P., 2016. Reliable ABC model choice via random forests. Bioinforma. Oxf. Engl. 32, 859–866. https://doi.org/10.1093/bioinformatics/btv684. Rossi, S., Pioz, M., Beard, E., Durand, B., Gibert, P., Gauthier, D., Klein, F., Maillard, D., Saint-Andrieux, C., Saubusse, T., Hars, J., 2014. Bluetongue dynamics in French wildlife: exploring the driving forces. Transbound. Emerg. Dis. 61, e12–24. https:// doi.org/10.1111/tbed.12061. Rushton, J., Lyons, N., 2015. Economic impact of Bluetongue: a review of the effects on production. Vet. Ital. 51, 401–406. https://doi.org/10.12834/VetIt.646.3183.1. Sailleau, C., Bréard, E., Viarouge, C., Vitour, D., Romey, A., Garnier, A., Fablet, A., Lowenski, S., Gorna, K., Caignard, G., Pagneux, C., Zientara, S., 2017. Re-emergence of bluetongue virus serotype 8 in France, 2015. Transbound. Emerg. Dis. 64, 998–1000. https://doi.org/10.1111/tbed.12453. Sedda, L., Brown, H.E., Purse, B.V., Burgin, L., Gloster, J., Rogers, D.J., 2012. A new algorithm quantifies the roles of wind and midge flight activity in the bluetongue epizootic in northwest Europe. Proc. Biol. Sci. 279, 2354–2362. https://doi.org/10. 1098/rspb.2011.2555. Sénat, 2008. Gestion de l’épizootie de fièvre catarrhale ovine (FCO) (No. 460), Rapport d’information. Singer, R.S., MacLachlan, N.J., Carpenter, T.E., 2001. Maximal predicted duration of viremia in bluetongue virus-infected cattle. J. Vet. Diagn. Investig. Off. Publ. Am. Assoc. Vet. Lab. Diagn. Inc. 13, 43–49. https://doi.org/10.1177/ 104063870101300109. Sumner, T., Orton, R.J., Green, D.M., Kao, R.R., Gubbins, S., 2017. Quantifying the roles of host movement and vector dispersal in the transmission of vector-borne diseases of livestock. PLoS Comput. Biol. 13, e1005470. https://doi.org/10.1371/journal.pcbi. 1005470. Szmaragd, C., Wilson, A.J., Carpenter, S., Wood, J.L.N., Mellor, P.S., Gubbins, S., 2009. A modeling framework to describe the transmission of bluetongue virus within and between farms in Great Britain. PLoS One 4, e7741. https://doi.org/10.1371/journal. pone.0007741. Tago, D., Hammitt, J.K., Thomas, A., Raboisson, D., 2014. Cost assessment of the movement restriction policy in France during the 2006 bluetongue virus episode (BTV-8). Prev. Vet. Med. 117, 577–589. https://doi.org/10.1016/j.prevetmed.2014. 10.010. Takamatsu, H.-H., Mellor, P.S., Mertens, P.P.C., 2004. A potential overwintering mechanism for bluetongue virus–recent findings. Vet. Ital. 40, 456–461. Turner, J., Bowers, R.G., Baylis, M., 2012. Modelling bluetongue virus transmission between farms using animal and vector movements. Sci. Rep. 2, 319. https://doi.org/ 10.1038/srep00319. Turner, J., Jones, A.E., Heath, A.E., Wardeh, M., Caminade, C., Kluiters, G., Bowers, R.G., Morse, A.P., Baylis, M., 2019. The effect of temperature, farm density and foot-andmouth disease restrictions on the 2007 UK bluetongue outbreak. Sci. Rep. 9. https:// doi.org/10.1038/s41598-018-35941-z. Vitour, D., Guillotin, J., Sailleau, C., Viarouge, C., Desprat, A., Wolff, F., Belbis, G., Durand, B., Bakkali-Kassimi, L., Breard, E., Zientara, S., Zanella, G., 2011. Colostral antibody induced interference of inactivated bluetongue serotype-8 vaccines in calves. Vet. Res. 42, 18. https://doi.org/10.1186/1297-9716-42-18. Wilson, A., Darpel, K., Mellor, P.S., 2008. Where does bluetongue virus sleep in the winter? PLoS Biol. 6. https://doi.org/10.1371/journal.pbio.0060210. Worwa, G., Hilbe, M., Chaignat, V., Hofmann, M.A., Griot, C., Ehrensperger, F., Doherr, M.G., Thür, B., 2010. Virological and pathological findings in Bluetongue virus serotype 8 infected sheep. Vet. Microbiol. 144, 264–273. https://doi.org/10.1016/j. vetmic.2010.01.011.

Flannery, J., Sanz-Bernardo, B., Ashby, M., Brown, H., Carpenter, S., Cooke, L., Corla, A., Frost, L., Gubbins, S., Hicks, H., Qureshi, M., Rajko-Nenow, P., Sanders, C., Tully, M., Bréard, E., Sailleau, C., Zientara, S., Darpel, K., Batten, C., 2019. Evidence of reduced viremia, pathogenicity and vector competence in a re-emerging European strain of bluetongue virus serotype 8 in sheep. Transbound. Emerg. Dis. https://doi.org/10. 1111/tbed.13131. Goffredo, M., Romeo, G., Monaco, F., Di Gennaro, A., Savini, G., 2004. Laboratory survival and blood feeding response of wild-caught Culicoides obsoletus Complex (Diptera: Ceratopogonidae) through natural and artificial membranes. Vet. Ital. 40, 282–285. Graesbøll, K., Bødker, R., Enøe, C., Christiansen, L.E., 2012. Simulating spread of Bluetongue Virus by flying vectors between hosts on pasture. Sci. Rep. 2, 863. https://doi.org/10.1038/srep00863. Græsbøll, K., Enøe, C., Bødker, R., Christiansen, L.E., 2014. Optimal vaccination strategies against vector-borne diseases. Spat. Spatiotemporal Epidemiol. 11, 153–162. https:// doi.org/10.1016/j.sste.2014.07.005. Gubbins, S., Szmaragd, C., Burgin, L., Wilson, A., Volkova, V., Gloster, J., Gunn, G.J., 2010. Assessing the consequences of an incursion of a vector-borne disease I. Identifying feasible incursion scenarios for bluetongue in Scotland. Epidemics 2, 148–154. https://doi.org/10.1016/j.epidem.2010.05.001. Hendrickx, G., Gilbert, M., Staubach, C., Elbers, A., Mintiens, K., Gerbier, G., Ducheyne, E., 2008. A wind density model to quantify the airborne spread of Culicoides species during north-western Europe bluetongue epidemic, 2006. Prev. Vet. Med. 87, 162–181. https://doi.org/10.1016/j.prevetmed.2008.06.009. Kluiters, G., Swales, H., Baylis, M., 2015. Local dispersal of palaearctic Culicoides biting midges estimated by mark-release-recapture. Parasit. Vectors 8, 86. https://doi.org/ 10.1186/s13071-015-0658-z. Lenormand, M., Jabot, F., Deffuant, G., 2013. Adaptive approximate Bayesian computation for complex models. Comput. Stat. 28, 2777–2796. https://doi.org/10.1007/ s00180-013-0428-3. Martinelle, L., Dal Pozzo, F., Sarradin, P., De Leeuw, I., De Clercq, K., Thys, C., Ziant, D., Thiry, E., Saegerman, C., 2011. Two alternative inocula to reproduce bluetongue virus serotype 8 disease in calves. Vaccine 29, 3600–3609. https://doi.org/10.1016/ j.vaccine.2011.02.055. Mellor, P.S., Carpenter, S., Harrup, L., Baylis, M., Mertens, P.P.C., 2008. Bluetongue in Europe and the Mediterranean Basin: history of occurrence prior to 2006. Prev. Vet. Med. 87, 4–20. https://doi.org/10.1016/j.prevetmed.2008.06.002. Mounaix, B., Caillaud, D., Echevarria, L., Reynaud, D., Fraboulet, M., Gorceix, M., Dupont, L., David, V., Lucbert, J., 2010. Estimation des impacts technicoéconomiques de la FCO-8 en 2007 au niveau de l’élevage. Bull. Épidémiol. 35, 17–19. Mullens, B.A., Gerry, A.C., Lysyk, T.J., Schmidtmann, E.T., 2004. Environmental effects on vector competence and virogenesis of bluetongue virus in Culicoides: interpreting laboratory data in a field context. Vet. Ital. 40, 160–166. Napp, S., Gubbins, S., Calistri, P., Allepuz, A., Alba, A., García-Bocanegra, I., Giovannini, A., Casal, J., 2011. Quantitative assessment of the probability of bluetongue virus overwintering by horizontal transmission: application to Germany. Vet. Res. 42, 4. https://doi.org/10.1186/1297-9716-42-4. Ninio, C., 2011. Fièvre catarrhale ovine dans les Ardennes : étude de la biologie des Culicoides et de leur rôle épidémiologique. Université Reims Champagne-Ardennes. Oura, C., Wood, J.L.N., Floyd, T., Sanders, A.J., Bin-Tarif, A., Henstock, M., Edwards, L., Simmons, H., Batten, C.A., 2010. Colostral antibody protection and interference with immunity in lambs born from sheep vaccinated with an inactivated Bluetongue serotype 8 vaccine. Vaccine 28, 2749–2753. https://doi.org/10.1016/j.vaccine.2010. 01.028. Palisson, A., Courcoul, A., Durand, B., 2017. Analysis of the spatial organization of pastures as a contact network, implications for potential disease spread and biosecurity in livestock, France, 2010. PLoS One 12, e0169881. https://doi.org/10.1371/journal. pone.0169881.

11